Trailers12k Dataset

Trailers12k is a movie trailer dataset comprised of 12,000 titles associated to ten genres. It distinguishes from other trailer datasets by its collection procedure aimed at providing a high-quality publicly available dataset.

This dataset is an updated and refined version of Trailers15k (Zenodo). In addition to manually-curated movie trailers, Trailers12k provides Kinetics video clip-level representations, ImageNet poster representations and rich meta-data.

If you find this work useful in your research, please consider citing.

@article{Trailers12k-2023103343,
title = {Improving Transfer Learning for Movie Trailer Genre Classification using a Dual Image and Video Transformer},
journal = {Information Processing & Management},
volume = {60},
number = {3},
pages = {103343},
year = {2023},
issn = {0306-4573},
doi = {https://doi.org/10.1016/j.ipm.2023.103343},
url = {https://www.sciencedirect.com/science/article/pii/S0306457323000808},
author = {Ricardo Montalvo-Lezama and Berenice Montalvo-Lezama and Gibran Fuentes-Pineda},
keywords = {Multi-label classification, Transfer learning, Trailers12k, Spatio-temporal analysis, Video analysis, Transformer model},
}